A format-preservation-based multi-document cross-consistency verification method and system

Abstract

The application provides a multi-document cross-consistency verification method and system based on format reservation, relates to the technical field of document intelligent review, and comprises the following steps: analyzing a bidding document, generating a plurality of review indexes, obtaining the importance levels of the review indexes, and calculating the correlation indexes between the review indexes; scoring each bidding document through the plurality of review indexes to generate index scores; combining the importance levels and the index scores to generate a first comprehensive score of each bidding document; correcting the first comprehensive score according to the correlation indexes to generate a second comprehensive score; performing text similarity analysis on the contents between the bidding documents to generate a second correction coefficient of each bidding document, correcting the second comprehensive score according to the second correction coefficient to generate a target comprehensive score, and sorting the bidding documents according to the target comprehensive score to generate a review result. The application has the technical effect of improving the accuracy of the scoring result.

Description

Technical Field

[0001] This application relates to the field of intelligent document review technology, specifically to a method and system for cross-consistency verification of multiple documents based on format preservation. Background Technology

[0002] In the field of bidding and procurement, the review process is of paramount importance, as its fairness and accuracy directly impact the quality of the bidding results. Traditional review processes rely heavily on expert experience; however, due to the large number of bid documents, numerous and interconnected review indicators, and the potential for plagiarism or borrowing among bid documents, manual review struggles to comprehensively control all dimensions, easily leading to inconsistent scoring or overlooking violations.

[0003] Currently, existing intelligent assisted review systems adopt rule-based scoring mechanisms, which automatically score bid documents through multiple preset scoring rules. Although this method improves review efficiency and enhances the objectivity of scoring, the scoring results are inaccurate because multiple preset scoring rules score independently during the scoring process and lack comprehensive correlation analysis. Summary of the Invention

[0004] This application provides a method and system for cross-validation of multiple documents based on format preservation, which can improve the accuracy of scoring results.

[0005] Firstly, this application provides a method for cross-validation of multiple documents based on format preservation. The method includes: acquiring a tender document; parsing the tender document to generate multiple evaluation indicators; obtaining the importance level of each evaluation indicator; and calculating the correlation index between the evaluation indicators; acquiring multiple bid documents corresponding to the tender document; scoring each bid document using the multiple evaluation indicators to generate an indicator score for each bid document corresponding to each evaluation indicator; generating weight coefficients for each evaluation indicator based on the importance level; and performing a weighted summation of each indicator score and its corresponding weight coefficient to generate a first comprehensive score for each bid document; and constructing each indicator based on the correlation index. A mapping matrix between scores is used to characterize the proportional relationship between the scores of each indicator. Based on the mapping matrix, a standard score corresponding to each indicator score is determined. Based on the difference between each indicator score and the corresponding standard score, a first correction coefficient is generated for each bid document. The first comprehensive score is corrected based on the first correction coefficient to generate a second comprehensive score. Text similarity analysis is performed on the content of each bid document. Based on the text similarity analysis results, a second correction coefficient is generated for each bid document. The second comprehensive score is corrected based on the second correction coefficient to generate a target comprehensive score. The bid documents are sorted based on the target comprehensive score to generate a review result.

[0006] By adopting the above technical solution, the system first extracts review indicators from the bidding documents and calculates the correlation index between the indicators, establishing the inherent relationship between indicator scores. Then, it determines weight coefficients based on importance levels to make the scores more aligned with actual needs. Next, it quantifies the ideal proportional relationship between indicator scores through a mapping matrix and generates a first correction coefficient using the difference between the standard score and the actual score, effectively correcting unreasonable deviations in the scoring results. Finally, it identifies duplicate content among bidding documents through text similarity analysis and uses a second correction coefficient to further refine the scores, thereby obtaining an accurate comprehensive target score. This multi-level scoring and correction mechanism ensures internal consistency in the scoring and improves the accuracy of duplicate content identification, making the final review results more objective and fair, and enhancing the accuracy of the scoring results.

[0007] Optionally, calculating the correlation index between each of the review indicators includes: obtaining preset business rules and review content of each of the review indicators; performing matching analysis on each of the review contents according to the business rules; when there is a logical relationship between any two of the review contents as defined by the business rules, calculating a business correlation index between the two review contents, wherein the business correlation index is positively correlated with the closeness of the logical relationship; and generating a correlation index between each of the review indicators based on the business correlation index, wherein the business correlation index is positively correlated with the correlation index.

[0008] By adopting the above technical solution and introducing preset business rules to match and analyze the review content, the system can identify potential logical relationships between review indicators and quantify the tightness of these relationships by calculating a business correlation index, ultimately generating an accurate correlation index between review indicators. This business rule-based correlation analysis method enables the system to deeply understand the intrinsic connections between review indicators from a business perspective, avoiding simple and mechanical independent scoring, providing a more reliable basis for subsequent score correction, and improving the rationality of the review results.

[0009] Optionally, constructing a mapping matrix between the scores of each indicator based on the correlation index includes: when the correlation index between two target review indicators is greater than a preset threshold, calculating a mapping coefficient between the scores of the two target review indicators based on the correlation index, wherein the mapping coefficient represents the ideal proportional relationship between the scores of the two corresponding indicators; recording the mapping coefficient in the mapping matrix, wherein the rows and columns of the mapping matrix correspond to each review indicator, and the elements in the mapping matrix represent the mapping coefficient between the scores of the two review indicators corresponding to the corresponding row and column; when the correlation index between two target review indicators is less than or equal to the preset threshold, setting the corresponding mapping coefficient to a preset value, wherein the preset value indicates that there is no correlation between the two corresponding review indicators.

[0010] By adopting the above technical solution and constructing a mapping matrix based on a threshold correlation index, the system achieves precise quantification of the correlation between review indicators. When the correlation index exceeds the threshold, the system calculates the corresponding mapping coefficient to characterize the ideal proportional relationship between indicator scores; when the correlation index does not exceed the threshold, a preset value is used to mark the uncorrelated state between indicators. This threshold-based mapping matrix construction method ensures the reasonable proportion of scores between strongly correlated indicators while avoiding excessive constraints on weakly correlated indicators, making score correction more accurate and controllable and enhancing the scientific nature of the review process.

[0011] Optionally, the step of generating a first correction coefficient for each bid document based on the difference between each indicator score and the corresponding standard score, and correcting the first comprehensive score based on the first correction coefficient to generate a second comprehensive score, includes: calculating the difference between each indicator score and the corresponding standard score; weighted summing of the absolute values ​​of each difference to generate a consistency deviation value for the bid document; calculating a first correction coefficient for each bid document based on the consistency deviation value, wherein the consistency deviation value is negatively correlated with the first correction coefficient; and arithmetically multiplying the first comprehensive score by the first correction coefficient to generate a second comprehensive score for the bid document.

[0012] By employing the aforementioned technical solution, the system calculates and weights the difference between the indicator score and the standard score, resulting in a consistency deviation value that reflects the overall degree of deviation in the scoring. Based on this deviation value, a first correction coefficient negatively correlated with the deviation is generated. This correction mechanism based on scoring deviation effectively reduces the scores of inconsistent bidding documents by correcting the first comprehensive score, guiding the review results closer to the standard score, improving the internal consistency of the scoring, and making the final second comprehensive score more accurate and reliable.

[0013] Optionally, the step of calculating the first correction coefficient for each of the bid documents based on the consistency deviation value includes: obtaining a preset deviation benchmark value and a decay coefficient, substituting the consistency deviation value, the deviation benchmark value, and the decay coefficient into a preset formula to calculate the first correction coefficient for each of the bid documents; wherein, the preset formula is: C=max(L, 1-max(0, DB)×λ); where C is the first correction coefficient, D is the consistency deviation value, B is the deviation benchmark value, λ is the decay coefficient, and L is a preset lower limit value of the correction coefficient.

[0014] By adopting the above technical solution and introducing a clear correction calculation formula, the system achieves accurate calculation of the first correction coefficient under the control of the deviation benchmark value and the attenuation coefficient. This formula uses a max function to ensure that the correction coefficient is not lower than the lower limit, uses the deviation benchmark value to set the correction initiation threshold, and uses the attenuation coefficient to control the correction intensity. This ensures that the correction process has both a reasonable margin of error and appropriate penalties for deviations exceeding the benchmark value. This correction mechanism based on a parameterized formula maintains the necessary flexibility and good controllability in the scoring correction, effectively improving the scientific rigor and reliability of the scoring correction.

[0015] Optionally, the step of generating a second correction coefficient for each of the bid documents based on the text similarity analysis results, and correcting the second comprehensive score based on the second correction coefficient to generate a target comprehensive score, includes: pairing each of the bid documents together, calculating the text similarity for each pair of bid documents to generate corresponding text similarity values, and preprocessing the images in each pair of bid documents to generate an image feature set; adjusting the text similarity values ​​based on the image feature set to generate a target text similarity value; generating a similarity aggregation index for each bid document based on each target text similarity value, wherein the similarity aggregation index represents the overall similarity between the bid document and other bid documents; calculating a second correction coefficient for each bid document based on the similarity aggregation index, wherein the similarity aggregation index is negatively correlated with the second correction coefficient; and arithmetically multiplying the second comprehensive score by the second correction coefficient to generate the target comprehensive score.

[0016] By employing the aforementioned technical solution and performing pairwise analysis on the bidding documents, the system not only calculates text similarity but also adjusts the similarity values ​​using image feature sets, generating a more comprehensive target text similarity value. Subsequently, based on these similarity values, it calculates a similarity aggregation index for each document and generates a negatively correlated second correction coefficient. This multi-dimensional similarity analysis method, combining text and image content, enables the system to more accurately identify substantial duplicate content between bidding documents. By correcting the second comprehensive score, it effectively reduces the final score of bidding documents exhibiting plagiarism or borrowing, thereby improving the fairness and accuracy of the review process.

[0017] Optionally, adjusting the text similarity value based on the image feature set to generate a target text similarity value includes: calculating the similarity value of images in each group of bidding documents based on the image feature set, wherein the similarity value is used to characterize the degree of consistency between images in two bidding documents; comparing the similarity value with a preset image similarity threshold, and marking the corresponding bidding document as having duplicate images when the similarity value is greater than the image similarity threshold; adjusting the text similarity value based on the marking result of duplicate images to generate a target text similarity value; wherein, when a bidding document is marked as having duplicate images, the target text similarity value is increased.

[0018] By adopting the above technical solution and introducing an image similarity analysis mechanism, the system calculates similarity values ​​based on image feature sets and sets image similarity thresholds for duplicate detection. When duplicate images are detected, the similarity value of the target text is increased accordingly. This method, which combines image analysis results with text similarity, overcomes the limitations of traditional methods that only focus on text content. It can effectively identify plagiarism through image copying, making similarity analysis more comprehensive and accurate. This provides a more reliable basis for subsequent scoring corrections and improves the ability to identify duplicate bidding documents.

[0019] Secondly, this application provides a multi-document cross-consistency verification system based on format preservation, the system comprising: a first acquisition module, a second acquisition module, a first generation module, a construction module, a second generation module, and a third generation module; wherein, The first acquisition module is used to acquire the bidding document, parse the bidding document, generate multiple evaluation indicators, acquire the importance level of each evaluation indicator, and calculate the correlation index between the evaluation indicators; the second acquisition module is used to acquire multiple bid documents corresponding to the bidding document, score each bid document using the multiple evaluation indicators, and generate indicator scores for each bid document corresponding to each evaluation indicator; the first generation module is used to generate weight coefficients for each evaluation indicator based on the importance level, and perform a weighted summation of each indicator score and its corresponding weight coefficient to generate a first comprehensive score for each bid document; the construction module is used to construct a mapping matrix between the indicator scores based on the correlation index. The mapping matrix is ​​used to characterize the proportional relationship between the scores of each indicator; the second generation module is used to determine the standard score corresponding to each indicator score according to the mapping matrix, generate a first correction coefficient for each bid document according to the difference between each indicator score and the corresponding standard score, correct the first comprehensive score according to the first correction coefficient, and generate a second comprehensive score; the third generation module is used to perform text similarity analysis on the content of each bid document, generate a second correction coefficient for each bid document according to the text similarity analysis results, correct the second comprehensive score according to the second correction coefficient, generate a target comprehensive score, sort the bid documents according to the target comprehensive score, and generate a review result.

[0020] Thirdly, this application provides an electronic device that adopts the following technical solution: it includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to execute a computer program of any of the above-described format-preserving multi-document cross-consistency verification methods.

[0021] Fourthly, this application provides a computer-readable storage medium that employs the following technical solution: storing a computer program capable of being loaded by a processor and executing any of the above-mentioned format-preserving multi-document cross-consistency verification methods.

[0022] In summary, this application includes at least one of the following beneficial technical effects: The system first extracts evaluation indicators from the bidding documents and calculates the correlation index between the indicators, establishing an inherent link between the indicator scores. Then, it determines weight coefficients based on importance levels to make the scores more aligned with actual needs. Next, it quantifies the ideal proportional relationship between indicator scores through a mapping matrix and generates a first correction coefficient using the difference between the standard score and the actual score, effectively correcting unreasonable deviations in the scoring results. Finally, it identifies duplicate content among the bidding documents through text similarity analysis and uses a second correction coefficient to further refine the scores, thus obtaining an accurate overall target score. This multi-layered scoring and correction mechanism ensures internal consistency in the scoring and improves the accuracy of duplicate content identification, making the final review results more objective and fair, and enhancing the accuracy of the scoring results. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a scenario provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a multi-document cross-consistency verification method based on format preservation provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a format-preserving multi-document cross-consistency verification system provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0024] Explanation of reference numerals in the attached figures: 1000, electronic device; 1001, processor; 1002, communication bus; 1003, user interface; 1004, network interface; 1005, memory. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] In the description of the embodiments in this application, words such as "illustrative," "for example," or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "illustrative," "for example," or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of words such as "illustrative," "for example," or "for example" is intended to present the relevant concepts in a specific manner.

[0027] Figure 1 This is a schematic diagram of a scenario provided in an embodiment of this application, such as... Figure 1 As shown in the diagram, this scenario illustrates the complete workflow of the format-preserving multi-document cross-validation method: the screen uses a three-column layout. The left side displays the three core documents input by the system (the tender document at the top and the two bid documents A and B at the bottom). The middle area presents the system's core analysis functions (including the weighting of review indicators for technical solutions (0.4), price (0.3), and service capabilities (0.3), the analysis of 67% text similarity between bid documents A and B and 75% similarity with the tender document, and the calculation of correction coefficients of 0.92 for bid documents A and 0.88 for bid documents B). The right side displays the final review results (the comprehensive score of bid document A is 87.5 points, the comprehensive score of bid document B is 82.3 points, and in the final ranking, document A ranks first and document B ranks second). The entire interface, through clear data visualization and a reasonable layout, intuitively demonstrates the complete review process from document input and multi-dimensional analysis to the final score ranking.

[0028] Figure 2 This is a flowchart illustrating a format-preserving multi-document cross-consistency verification method provided in an embodiment of this application. Figure 2 As shown, the method includes S101-S106: S101: Obtain the bidding documents, parse the bidding documents, generate multiple evaluation indicators, obtain the importance level of each evaluation indicator, and calculate the correlation index between each evaluation indicator.

[0029] In bidding review practice, to ensure the objectivity and scientific nature of the review process, it is necessary to systematically analyze the bidding documents in order to establish a standardized review system.

[0030] In practice, the bidding documents are first analyzed in a structured manner using a document parsing module. A combination of keyword matching and semantic analysis is employed to extract evaluation indicators, such as "completeness of technical solution," "reasonableness of price," and "service responsiveness." Then, based on historical data from the evaluation experience database, each evaluation indicator is assigned an importance level, categorized into three levels: "critical," "important," and "general." "Critical" indicators directly affect bidding eligibility, "important" indicators have a significant impact on scoring, and "general" indicators serve as a scoring reference. Based on the determined importance levels, a correlation index is calculated by analyzing the business relationships between the evaluation indicators. The correlation index ranges from [0,1], representing the degree of correlation between two evaluation indicators. For example, the correlation index between "completeness of technical solution" and "service responsiveness" might be 0.7, indicating a strong business correlation between the two indicators.

[0031] Taking a government agency's procurement project for a smart city monitoring system as an example, the project budget is 10 million yuan. The system first analyzes the document through a document parsing module: extracting basic qualification requirements from the "Bidder Qualification Requirements" section, such as "possessing a value-added telecommunications business operating license"; extracting technical indicator requirements from the "Technical Requirements" section, such as "video analysis accuracy rate not less than 98%"; and extracting service indicator requirements from the "Business Requirements" section, such as "fault response time not exceeding 30 minutes". After system analysis and manual verification, 12 core review indicators are finally generated: 6 technical indicators (including system architecture design, data processing capability, algorithm performance indicators, system scalability, security measures, and interface standardization), 4 business indicators (including bid price, performance capability, after-sales service, and implementation cycle), and 2 qualification indicators (including enterprise qualifications and project performance). Based on industry standards and historical bidding experience, the system classifies these indicators into importance levels: "System Architecture Design", "Data Processing Capability", and "Bidding Price" are classified as "Critical", with weightings of 20%, 15%, and 25%, respectively; "Algorithm Performance Indicators", "Security Measures", and "After-Sales Service" are classified as "Important", with weightings of 10%, 10%, and 8%, respectively; and the remaining indicators are classified as "General", with weightings between 2% and 5%.

[0032] When calculating the correlation index, the system found that the correlation index between "system architecture design" and "data processing capability" was 0.8, indicating a high correlation between the two indicators; while the correlation index between "bid price" and "project performance" was only 0.3, indicating a low correlation between the two. This refined indicator system design allows subsequent review work to more objectively assess the overall strength of the bidding proposal, avoiding the situation where a single indicator excessively dominates the review results. At the same time, the introduction of the correlation index also provides important evidence for identifying unreasonable pricing or technical solutions in the bid documents.

[0033] Based on the above embodiments, as an optional implementation method, in S101, calculating the correlation index between each review indicator specifically includes S11-S13: S11, obtain the preset business rules and the review content of each review indicator.

[0034] In the correlation analysis of review indicators for smart city monitoring system projects, to accurately quantify the relationships between review indicators, the system first retrieves pre-defined business rules and review content from the review knowledge base. Business rules include rule types such as "technology dependency," "functional implementation constraints," and "resource matching requirements," each defining a specific logical relationship pattern. For example, the "technology dependency" rule stipulates that if the implementation of indicator A depends on the technical support provided by indicator B, then the two are strongly correlated; the "functional implementation constraints" rule stipulates that if the functional implementation of indicator A affects the performance of indicator B, then the two are moderately correlated; and the "resource matching requirements" rule stipulates that if indicator A and indicator B share the same resource pool, then the two are weakly correlated. Simultaneously, the system extracts the specific review content for each review indicator. For example, the review content for "system architecture design" includes "distributed architecture design scheme," "system fault tolerance mechanism," and "extension interface definition," while the review content for "data processing capability" includes "real-time data throughput," "data processing latency," and "concurrency processing mechanism."

[0035] S12, perform matching analysis on each review content according to business rules. When there is a logical relationship between any two review contents as defined by the business rules, calculate the business relationship index between the two review contents. The business relationship index is positively correlated with the closeness of the logical relationship.

[0036] S13. Based on the business correlation index, generate the correlation index between each review indicator. The business correlation index and the correlation index are positively correlated.

[0037] The system automatically matches and analyzes the review content according to business rules, finding a logical relationship of "technology dependency" between "distributed architecture design scheme" and "real-time data throughput," as the distributed architecture directly determines the system's data processing capability. Based on this, the business correlation index for these two review items is calculated to be 0.9. A logical relationship of "functional implementation constraints" is found between "system fault tolerance mechanism" and "data processing latency," resulting in a business correlation index of 0.6. A logical relationship of "resource matching requirements" is found between "extended interface definition" and "concurrent processing mechanism," resulting in a business correlation index of 0.3. After obtaining the business correlation indices at the review content level, the system calculates the correlation index at the review indicator level using a weighted average method, with weights determined based on the importance of the review content within the corresponding indicator. The final correlation index for "system architecture design" and "data processing capability" is 0.8 (a weighted average of the business correlation indices of multiple review items), indicating a strong correlation between these two review indicators.

[0038] S102: Obtain multiple bid documents corresponding to the bidding document, score each bid document using multiple evaluation indicators, and generate the indicator score for each bid document corresponding to each evaluation indicator.

[0039] In the bidding review process, in order to ensure the accuracy of the evaluation of the bidding documents and the objectivity and fairness of the scoring, it is necessary to establish a standardized indicator scoring system.

[0040] In this embodiment, the system obtains three bid documents corresponding to the tender documents, from bidding companies A, B, and C respectively. The scoring process combines rule-based automatic scoring with expert review to quantitatively evaluate various evaluation indicators for each bid document. The system first refines the evaluation indicators into quantifiable scoring items, such as subdividing "system architecture design" into "architectural integrity (30 points)," "technological advancement (40 points)," and "implementation feasibility (30 points)"; and "data processing capability" into "concurrency (40 points)," "response time (30 points)," and "accuracy (30 points)." For these specific scoring items, the system automatically scores by extracting key parameters from the bid documents. For example, it is found that Company A's bid document promises "video analysis accuracy of 99.2%," which exceeds the tender requirement of 98%, earning it 28 points in the "accuracy" scoring item; Company B's promise of "fault response time of 15 minutes" is better than the tender requirement of 30 minutes, earning it 27 points in the "response time" scoring item. For scoring items that require professional judgment, such as "technological advancement", the system will push relevant content to expert judges for scoring and ensure the rationality of the scoring through mechanisms such as scoring deviation verification.

[0041] Ultimately, the system generates a complete score sheet for each tender document. For example, Company A scored 92 points in the "System Architecture Design" indicator and 88 points in the "Data Processing Capability" indicator. This refined scoring mechanism not only improves the accuracy and interpretability of the scores but also provides reliable basic data for subsequent comprehensive score calculations, making the entire review process more transparent and standardized.

[0042] For example, in a smart city monitoring system bidding project, the system's evaluation of Company A's bid documents showed the following: In the technical categories, the system architecture design scored 92 points, demonstrating the advanced nature and completeness of its solution; the data processing capability scored 88 points, reflecting its technical strength; and the algorithm performance scored 85 points, indicating that its algorithm solution was basically reliable. In the business categories, the bid price scored 90 points (the price was reasonably reasonable), the performance capability scored 87 points (rich experience in past projects), and the after-sales service scored 95 points (detailed service plan and short response time). This multi-dimensional evaluation not only comprehensively reflects the merits of the bid solution but also provides an objective basis for subsequent comprehensive evaluation.

[0043] S103. Based on the importance level, generate the weight coefficients for each review indicator, and sum the scores of each indicator with their corresponding weight coefficients to generate the first comprehensive score for each bid document.

[0044] In bidding review practice, different review indicators have significantly different degrees of importance to bidding proposals. It is necessary to use a scientific weighting mechanism to convert the importance level into specific weight coefficients, so as to ensure that the scoring results can accurately reflect the comprehensive strength of each bidding proposal.

[0045] This embodiment employs a weight coefficient generation method based on importance levels. First, the importance level of the evaluation indicators is quantified into a weight benchmark value. The weight benchmark value for "critical" level indicators is set to 0.5, for "important" level indicators to 0.3, and for "general" level indicators to 0.2. Taking a smart city monitoring system bidding project as an example, the system normalizes the weights based on the weight benchmark values ​​and the number of evaluation indicators to ultimately determine the weight coefficients for each indicator: Among technical indicators, system architecture design (critical level) has a weight of 0.20, data processing capability (critical level) has a weight of 0.15, and algorithm performance indicators (important level) have a weight of 0.10; among business indicators, bid price (critical level) has a weight of 0.25, contract performance capability (important level) has a weight of 0.08, and after-sales service (important level) has a weight of 0.08; the remaining general level indicators have weights between 0.02 and 0.05.

[0046] After determining the weighting coefficients, the system multiplies Company A's scores for each indicator by their corresponding weighting coefficients and sums them. The calculation process is as follows: 92 × 0.20 (system architecture design) + 88 × 0.15 (data processing capability) + 85 × 0.10 (algorithm performance indicators) + 90 × 0.25 (bid price) + 87 × 0.08 (performance capability) + 95 × 0.08 (after-sales service) + weighted scores of other indicators. The final comprehensive score for Company A is 89.6. The same method is used to calculate the comprehensive score for Company B, which is 85.3, and for Company C, which is 82.7.

[0047] S104. Based on the correlation index, construct a mapping matrix between the scores of each indicator. The mapping matrix is ​​used to represent the proportional relationship between the scores of each indicator.

[0048] In the bidding and tendering process, there are often inherent logical connections and constraints between various evaluation indicators. In order to identify and quantify these relationships, it is necessary to construct a mapping matrix between indicator scores.

[0049] In the smart city monitoring system project, the system first sets a threshold of 0.6 for constructing the mapping matrix based on the previously calculated correlation index. When the correlation index between two evaluation indicators exceeds this threshold, the system calculates the mapping coefficient between the scores of these two indicators. Taking "system architecture design" and "data processing capability" as examples, their correlation index is 0.8, exceeding the threshold of 0.6. By analyzing historical bidding cases and industry standards, it is determined that the reasonable scoring ratio between these two indicators should be 1.2:1, meaning the system architecture design score should be slightly higher than the data processing capability score. This ratio is recorded as the mapping coefficient 1.2. Similarly, the correlation index between "algorithm performance indicator" and "data processing capability" is 0.75, and their mapping coefficient is set to 0.9, indicating that the algorithm performance indicator score should be slightly lower than the data processing capability score. However, the correlation index between "bid price" and "performance capability" is 0.3, below the threshold. The system sets the corresponding mapping coefficient to the preset value of 1, indicating that there is no significant proportional relationship between these two indicators. The final generated mapping matrix is ​​a 12×12 square matrix (corresponding to 12 evaluation indicators), and the element aij in the matrix represents the ideal ratio between the score of the i-th indicator and the score of the j-th indicator.

[0050] For example, in Company A's bid document, the actual ratio of the system architecture design score of 92 points to the data processing capability score of 88 points is 1.05, which deviates from the mapping coefficient of 1.2. This deviation will be taken into account in subsequent scoring corrections. By constructing such a mapping matrix, the system can evaluate the rationality of the scores for each indicator from an overall perspective, identify potential inconsistencies in the scoring, and provide an important basis for subsequent scoring corrections, thereby improving the reliability and consistency of the review results.

[0051] Based on the above embodiments, as an optional implementation, in S104, constructing the mapping matrix between the scores of each indicator according to the correlation index specifically includes S41-S42: S41, when the correlation index between two target evaluation indicators is greater than a preset threshold, the mapping coefficient between the indicator scores corresponding to the two target evaluation indicators is calculated based on the correlation index. The mapping coefficient represents the ideal proportional relationship between the corresponding two indicator scores.

[0052] In the review process of smart city monitoring system projects, in order to accurately depict the scoring mapping relationship between review indicators, the system needs to construct a scientific and reasonable mapping matrix. First, the system sets the correlation index threshold to 0.6. When the correlation index between two review indicators exceeds this threshold, it indicates that they have a significant business correlation, and the corresponding mapping coefficient needs to be calculated. Taking "system architecture design" and "data processing capability" as examples, the correlation index between the two is 0.8, which exceeds the preset threshold of 0.6. Based on historical bidding data and industry experience, the system constructs a mapping coefficient calculation model: when the correlation index is in the range of [0.6, 0.7], the mapping coefficient range is [0.8, 1.2]; when the correlation index is in the range of [0.7, 0.8], the mapping coefficient range is [0.7, 1.3]; when the correlation index is in the range of [0.8, 1.0], the mapping coefficient range is [0.6, 1.5].

[0053] S42, record the mapping coefficients into the mapping matrix, where the rows and columns of the mapping matrix correspond to each review indicator, and the elements in the mapping matrix represent the mapping coefficients between the indicator scores of the two review indicators corresponding to the corresponding row and column; when the correlation index between two target review indicators is less than or equal to a preset threshold, the corresponding mapping coefficient is set to a preset value, which indicates that there is no correlation between the two review indicators.

[0054] In the specific calculation, the system also considered the difference in importance levels between the two indicators: if the two indicators have the same importance level, the mapping coefficient is in the middle of its range; if there is a difference in importance level, the mapping coefficient shifts towards the higher-level indicator. Therefore, since both "system architecture design" and "data processing capability" are key indicators with a correlation index of 0.8, the system calculated a mapping coefficient of 1.2 between them, indicating that under a reasonable scoring condition, the score for system architecture design should be 1.2 times the score for data processing capability. The system filled the calculated mapping coefficient into the corresponding positions in the mapping matrix, constructing a 12×12 square matrix (corresponding to 12 review indicators).

[0055] In the matrix, all diagonal elements are 1 (representing the same indicator). For off-diagonal elements, if the correlation index between two corresponding indicators is less than or equal to the threshold of 0.6, the mapping coefficient is set to a preset value of 1, indicating that there is no significant scoring constraint between the two indicators. For example, the correlation index between "bid price" and "project performance" is 0.3, which is lower than the threshold of 0.6, so their corresponding mapping coefficient in the mapping matrix is ​​set to 1. This mapping matrix construction method based on correlation strength ensures the rationality of scoring between strongly correlated indicators while avoiding excessive constraints, making score correction more accurate. In this project, guided by the mapping matrix, the system successfully identified that the ratio of "system architecture design" score (92 points) to "data processing capability" score (88 points) in Company A's bid document, at 1.05, deviated from the ideal ratio of 1.2. This finding provided an important reference for subsequent score correction.

[0056] S105. Based on the mapping matrix, determine the standard score corresponding to each indicator score. Based on the difference between each indicator score and the corresponding standard score, generate the first correction coefficient for each bid document. Based on the first correction coefficient, correct the first comprehensive score to generate the second comprehensive score.

[0057] In the bidding and tendering process, to ensure the rationality and consistency of the scoring results, it is necessary to correct the initial scores based on the mapping matrix. In the smart city monitoring system project, the system first uses the mapping matrix to determine the standard scores for each indicator. The calculation method is to select the indicator pair with the highest correlation index as the benchmark and substitute its actual score into the mapping relationship for derivation. Taking Company A as an example, the system architecture design score of 92 points is used as the benchmark score. According to the mapping coefficient of 1.2, the standard score for data processing capability should be 76.7 points (92 / 1.2), while the actual score is 88 points, with a difference of 11.3 points. Similarly, the standard score for the algorithm performance indicator should be 85.2 points (76.7×0.9), while the actual score is 85 points, with a difference of -0.2 points. The system performs a weighted sum of the differences between all indicator scores and the standard scores, using the previously determined indicator weight coefficients, and obtains a consistency deviation value of 8.6 for Company A's tender documents.

[0058] According to the preset scoring correction rules, the system uses a deviation benchmark value B=5, a decay coefficient λ=0.02, and a correction coefficient lower limit L=0.8. These parameters are substituted into the formula C=max(L, 1-max(0, DB)×λ) to calculate the first correction coefficient, where D is the consistency deviation value. Substituting the data, the first correction coefficient for Company A is 0.93, indicating that the internal consistency of its various indicator scores is acceptable but still needs to be moderately reduced. Multiplying Company A's first comprehensive score of 89.6 points by the first correction coefficient 0.93 yields a second comprehensive score of 83.3 points. Similarly, the consistency deviation value for Company B is calculated to be 12.4, the first correction coefficient is 0.85, and the second comprehensive score is 72.5 points; for Company C, the consistency deviation value is 6.2, the first correction coefficient is 0.98, and the second comprehensive score is 81.0 points. Through this scoring correction mechanism based on a mapping matrix, the system can effectively identify and adjust unreasonable phenomena in the scoring, making the final score more objective and fair, and providing important decision-making references for review experts.

[0059] The basis for determining the deviation benchmark value B=5 is as follows: Analysis of the scoring data of 1000 historical high-quality bid documents revealed that, under reasonable evaluation indicator configuration, approximately 85% of the high-quality bid documents had a consistency deviation value within 5 points. This indicates that when the consistency deviation value of the indicator scores does not exceed 5 points, the scores of each indicator of the bid document are basically at a reasonable level and do not require correction. Therefore, the value of B is set to 5 so that when the deviation value is less than or equal to 5, the correction coefficient remains at 1, and no adjustment is made to the original score.

[0060] The determination of the attenuation coefficient λ=0.02 is based on the fact that this parameter determines the severity of the penalty after exceeding the deviation benchmark. Through fitting analysis of historical data, the correction effect is most ideal when λ=0.02: taking a consistency deviation of 15 points as an example, substituting into the formula yields a correction coefficient of 0.8 (1-max(0,15-5)×0.02=0.8), meaning a 20% reduction in the original score. This adjustment effectively suppresses score imbalance without causing excessive penalty. If λ=0.01 is chosen, the penalty is too weak and cannot effectively correct the score deviation; if λ=0.03 is chosen, it may lead to over-correction, distorting the scoring results.

[0061] The basis for determining the lower limit of the correction coefficient, L=0.8, is as follows: In actual review, even if the bidding documents have significant deviations in indicator scoring, their basic score should be retained to reflect their fundamental service capabilities. Analysis of historical data on bidding documents judged as "qualified but requiring close attention" revealed that lowering their scores by 20% (i.e., retaining 80%) provides sufficient punishment without completely disqualifying them from bidding. Therefore, the value of L is set to 0.8 to ensure that the corrected score will not fall below 80% of the original score.

[0062] Based on the above embodiments, as an optional implementation method, in S105, a first correction coefficient is generated for each bid document according to the difference between each indicator score and the corresponding standard score. The first comprehensive score is then corrected according to the first correction coefficient to generate a second comprehensive score, specifically including S51-S53: S51, calculate the difference between each indicator score and the corresponding standard score, and sum the absolute values ​​of each difference in a weighted manner to generate the consistency deviation value of the tender document.

[0063] During the review process of smart city monitoring system projects, to ensure the internal consistency of the scoring results, the system needs to quantitatively analyze and correct the deviation between the indicator scores and the standard scores. Taking Company A's bidding document as an example, the system first calculates the difference between the scores of each indicator and their standard scores: the actual score for system architecture design is 92 points, while the standard score derived from the mapping relationship is 90 points, a difference of 2 points; the actual score for data processing capability is 88 points, while the standard score is 76.7 points (derived from the system architecture design score and the mapping coefficient 1.2), a difference of 11.3 points; the actual score for algorithm performance is 85 points, while the standard score is 85.2 points (derived from the data processing capability standard score and the mapping coefficient 0.9), a difference of -0.2 points; the differences for other indicators are calculated in the same manner. Considering the varying importance of different indicators, the system uses pre-determined indicator weights to weight and sum the absolute values ​​of these differences: 2 × 0.20 (system architecture design weight) + 11.3 × 0.15 (data processing capability weight) + 0.2 × 0.10 (algorithm performance indicator weight) + other weighted differences, ultimately yielding a consistency deviation value of 8.6 for Company A's bid document. This value indicates a certain degree of inconsistency among the scores of various indicators in Company A's bid document.

[0064] S52, Calculate the first correction factor for each tender document based on the consistency deviation value, wherein the consistency deviation value is negatively correlated with the first correction factor.

[0065] S53, multiply the first comprehensive score by the first correction coefficient arithmetically to generate the second comprehensive score of the tender document.

[0066] Based on the consistency deviation value, the system calculates the first correction coefficient using a preset correction formula: C = max(0.8, 1 - max(0, 8.6 - 5) × 0.02) = 0.93, where 0.8 is the lower limit of the correction coefficient, 5 is the deviation baseline value, and 0.02 is the attenuation coefficient. This correction coefficient reflects the system's penalty for scoring inconsistencies; the larger the consistency deviation value, the smaller the correction coefficient. Finally, the system multiplies Company A's first comprehensive score of 89.6 points by the first correction coefficient of 0.93 to obtain the second comprehensive score of 83.3 points. Similarly, Company B's consistency deviation value is calculated to be 12.4, corresponding to a first correction coefficient of 0.85, and a second comprehensive score of 72.5 points (85.3 × 0.85); Company C's consistency deviation value is 6.2, corresponding to a first correction coefficient of 0.98, and a second comprehensive score of 81.0 points (82.7 × 0.98). This scoring correction mechanism based on consistency deviation effectively reduces the scores of inconsistent bidding documents, guiding bidding units to provide more reasonable technical solutions. In this case, although Company A scored highly on each individual indicator, the score decreased after correction due to obvious inconsistencies between the indicators. This result is more in line with the objectivity requirements of the review.

[0067] S106. Perform text similarity analysis on the content of each bidding document. Based on the text similarity analysis results, generate a second correction coefficient for each bidding document. Adjust the second comprehensive score based on the second correction coefficient to generate a target comprehensive score. Sort the bidding documents based on the target comprehensive score to generate the review results.

[0068] In bidding and tendering review practices, in order to prevent plagiarism, similarity, and other violations in bidding documents, it is necessary to use text similarity analysis for identification and scoring correction.

[0069] In the smart city monitoring system project, the system first paired off three bidding documents (A, B, and C) and calculated their text similarity. The calculation process employed a semantic analysis-based text similarity algorithm, segmenting the document content by paragraph, extracting keywords and semantic features, and calculating the similarity using methods such as cosine similarity. Simultaneously, the system preprocessed the images in the documents, generating image feature sets through image feature extraction technology, including information such as image size, color distribution, and texture features. Initially, the text similarity between documents A and B was 45%. After discovering a 90% similarity in the system architecture diagrams between the two documents (exceeding the preset image similarity threshold of 80%), the system adjusted the text similarity value to 55%. The similarity between documents B and C was 38%, with the image similarity not exceeding the threshold, so the original value was maintained. Similarly, the similarity between documents A and C was 32%, with the image similarity not exceeding the threshold, so the original value was also maintained.

[0070] Based on the adjusted similarity values, the system calculates a similarity aggregation index for each document. The calculation method is a weighted average of the similarity values ​​of the document and all other documents, with the weights proportional to the similarity values. Document A has a similarity aggregation index of 0.48 (mainly influenced by its high similarity to document B), document B has 0.42, and document C has 0.35. The system then substitutes the similarity aggregation index into a preset correction formula to calculate a second correction coefficient. The formula uses a decreasing function, meaning that the higher the similarity, the greater the correction. The final second correction coefficients are 0.90 for document A, 0.92 for document B, and 0.95 for document C. Multiplying the second comprehensive score of each bidding document by its corresponding second correction coefficient yields a target comprehensive score of 75.0 (83.3 × 0.90) for document A, 66.7 (72.5 × 0.92) for document B, and 77.0 (81.0 × 0.95) for document C. Based on the overall target score, the final review results were obtained: Document C ranked first, Document A ranked second, and Document B ranked third.

[0071] This secondary correction mechanism based on text similarity effectively reduces the scores of identical documents, guides bidding units to provide original solutions, and improves the fairness and credibility of the review results. In this case, although document A had the highest second comprehensive score, it was surpassed by document C after the second correction due to its high similarity to document B. This result is more in line with the principles of fairness and impartiality in bidding.

[0072] Based on the above embodiments, as an optional implementation method, in S106, a second correction coefficient is generated for each bidding document according to the text similarity analysis results, and the second comprehensive score is corrected according to the second correction coefficient to generate the target comprehensive score, specifically including S61-S65: S61, combine each bidding document in pairs, calculate the text similarity of each pair of bidding documents to generate the corresponding text similarity value, and preprocess the images in each pair of bidding documents to generate an image feature set.

[0073] During the review process of smart city monitoring system projects, to prevent plagiarism or similarity in bidding documents, the system needs to conduct in-depth screening through text similarity analysis. First, the system pairs three bidding documents (A, B, and C) and calculates similarity using a semantic analysis-based algorithm. The algorithm first segments the documents into words and extracts keywords, then calculates word frequency feature vectors using the TF-IDF method, and finally calculates text similarity using the cosine similarity formula. Simultaneously, the system extracts features from the image content in the documents, including image size ratio, color distribution, texture features, edge features, etc., generating standardized feature vectors.

[0074] S62, adjust the text similarity value based on the image feature set, and generate the target text similarity value.

[0075] Calculations showed that the initial text similarity between documents A and B was 45%. Image feature comparison revealed a 90% similarity in the system architecture diagrams between the two documents (exceeding the preset image similarity threshold of 80%). Based on this, the system increased the text similarity value by 10 percentage points to 55%. The similarity between documents B and C was 38%, and image feature comparison did not find significant similarity (the highest similarity was 65%, which did not exceed the threshold), so the original value was maintained. The similarity between documents A and C was 32%, and the image similarity also did not exceed the threshold, so the original value was maintained.

[0076] Based on the above embodiments, as an optional implementation, in S62, adjusting the text similarity value according to the image feature set to generate the target text similarity value specifically includes S621-S623: S621, Based on the image feature set, calculate the similarity value of the images in each group of bidding documents. The similarity value is used to characterize the degree of consistency between the images in the two bidding documents.

[0077] During the review process of the smart city monitoring system project, in order to accurately identify the similarity of images in the bidding documents, the system needs to conduct a detailed analysis of the image content in the documents. First, the system extracts features from the images in the bidding documents of Company A and Company B, generating image feature sets. The feature sets include three basic types of information: image size and aspect ratio, overall layout structure of the image, and text annotations in the image. Taking the system architecture diagram as an example, the system extracts feature information such as image pixel size (1920×1080), module layout (4-layer architecture, number and position of modules in each layer), connection relationships (connection methods between modules), and text annotations (module name, descriptive text).

[0078] S622, compare the similarity value with the preset image similarity threshold. When the similarity value is greater than the image similarity threshold, mark the corresponding bidding document as having duplicate images.

[0079] After obtaining the feature set, the system calculates the similarity value of the corresponding images in each group of bidding documents. The calculation method compares the features of the two images one by one: first, the image size is compared, and if the ratio difference is within 5%, it is considered similar; second, the layout structure is compared, including the number of modules, their position distribution, and connection methods, and if the proportion of identical elements exceeds 80%, it is considered similar; finally, the text annotations are compared, and the consistency of the text content is calculated using character matching. The system combines the comparison results of these three aspects to obtain the final image similarity value. For example, if the system architecture diagrams of Company A and Company B are completely identical in size ratio, the similarity of their layout structure is 92% (with only a few module position adjustments), and the consistency of their text annotations reaches 88%, the overall similarity value is calculated to be 90%.

[0080] S623, Based on the results of the image duplication marking, adjust the text similarity value and generate the target text similarity value; wherein, when the tender document is marked as having image duplication, the target text similarity value is increased.

[0081] The calculated similarity values ​​are compared with a preset image similarity threshold of 80%. Since the system architecture diagrams of documents A and B have a similarity (90%) exceeding the threshold, the system marks these documents as having "image duplication." The highest image similarity between documents B and C is 65%, and the highest image similarity between documents A and C is 58%, both below the threshold, therefore no marking is made.

[0082] Based on the image duplication flags, the system adjusts the text similarity values. The adjustment rule is as follows: when a document is flagged for image duplication, the original text similarity value is increased by 10 percentage points. Therefore, the text similarity between documents A and B is adjusted from 45% to 55% (target text similarity value); the similarity between documents B and C remains unchanged at 38%; and the similarity between documents A and C remains unchanged at 32%. This adjustment method ensures a reasonable penalty for image duplication while maintaining the overall balance of the similarity assessment. In this way, the system can effectively identify and handle image duplication in bidding documents, providing a reliable basis for subsequent scoring corrections.

[0083] S63. Based on the similarity values ​​of each target text, generate a similarity aggregation index for each bid document. The similarity aggregation index represents the overall similarity between the bid document and other bid documents.

[0084] S64. Calculate the second correction coefficient for each tender document based on the similarity aggregation index, wherein the similarity aggregation index and the second correction coefficient are negatively correlated.

[0085] S65, multiply the second comprehensive score by the second correction coefficient arithmetically to generate the target comprehensive score.

[0086] After obtaining the adjusted target text similarity values, the system calculates the similarity aggregation index for each bidding document. The calculation method involves a weighted average of the similarity values ​​of that document with all other documents, with the weights proportional to the similarity values. Specifically, if the similarity between two documents is x, then the weight of that similarity in calculating the aggregation index is x². This weighting emphasizes the impact of high similarity. The calculated similarity aggregation index for document A is 0.48 (mainly influenced by its 55% similarity to document B), for document B it is 0.42, and for document C it is 0.35. Based on the similarity aggregation index, the system uses a preset correction formula to calculate the second correction coefficient: M = 1 - k × I², where k is the adjustment coefficient (value 0.5), and I is the similarity aggregation index. Substituting the data, the second correction coefficient for document A is 0.90, for document B it is 0.92, and for document C it is 0.95. Finally, the system multiplies the second comprehensive score of each tender document by the corresponding second correction coefficient, resulting in a target comprehensive score of 75.0 points (83.3×0.90) for document A, 66.7 points (72.5×0.92) for document B, and 77.0 points (81.0×0.95) for document C.

[0087] This multi-level correction mechanism based on text similarity not only considers the similarity of text content but also the degree of similarity of non-text content such as images. It effectively identifies bidding documents with high overall similarity through the calculation of a similarity aggregation index. In this case, although document A had a high second-level comprehensive score, its score was significantly reduced after the second correction due to its high similarity to document B. This result is more conducive to maintaining the fairness and innovativeness of the bidding process.

[0088] Based on the above method, this application also discloses a format-preserving multi-document cross-consistency verification system, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of a format-preserving multi-document cross-consistency verification system provided in an embodiment of this application. The system includes: a first acquisition module, a second acquisition module, a first generation module, a construction module, a second generation module, and a third generation module; wherein, The first acquisition module acquires the bidding documents, parses them, generates multiple evaluation indicators, obtains the importance level of each indicator, and calculates the correlation index between the indicators. The second acquisition module acquires multiple bid documents corresponding to the bidding documents, scores each bid document using the multiple evaluation indicators, and generates indicator scores for each bid document corresponding to each evaluation indicator. The first generation module generates weight coefficients for each evaluation indicator based on its importance level, and performs a weighted summation of each indicator score and its corresponding weight coefficient to generate a first comprehensive score for each bid document. The construction module constructs a mapping matrix between the indicator scores based on the correlation index. The mapping matrix is ​​used to represent the proportional relationship between the scores of each indicator; the second generation module is used to determine the standard score corresponding to each indicator score according to the mapping matrix, generate the first correction coefficient for each bid document based on the difference between each indicator score and the corresponding standard score, correct the first comprehensive score according to the first correction coefficient, and generate the second comprehensive score; the third generation module is used to perform text similarity analysis on the content of each bid document, generate the second correction coefficient for each bid document based on the text similarity analysis results, correct the second comprehensive score according to the second correction coefficient, generate the target comprehensive score, sort the bid documents according to the target comprehensive score, and generate the review results.

[0089] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0090] Please see Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0091] The communication bus 1002 is used to realize the connection and communication between these components.

[0092] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0093] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0094] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 1001 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 1001 and may be implemented as a separate chip.

[0095] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 4 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program based on a format-preserving multi-document cross-consistency verification method.

[0096] exist Figure 4 In the electronic device 1000 shown, the user interface 1003 is mainly used to provide an interface for users to input data and obtain user input data; while the processor 1001 can be used to call an application stored in the memory 1005 that is a format-preserving multi-document cross-consistency verification method. When executed by one or more processors, the electronic device performs one or more of the methods described in the above embodiments.

[0097] An electronic device readable storage medium stores instructions that, when executed by one or more processors, cause the electronic device to perform one or more of the methods described in the above embodiments.

[0098] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0099] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0100] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some service interfaces; indirect couplings or communication connections between devices or units may be electrical or other forms.

[0101] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0102] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0103] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0104] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for cross-validation of multiple documents based on format preservation, characterized in that, The method includes: Obtain the bidding documents, parse the bidding documents, generate multiple evaluation indicators, obtain the importance level of each evaluation indicator, and calculate the correlation index between each evaluation indicator. Obtain multiple bid documents corresponding to the bidding document, score each bid document using multiple evaluation indicators, and generate indicator scores for each bid document corresponding to each evaluation indicator. Based on the importance level, a weight coefficient for each of the evaluation indicators is generated. The scores of each indicator are weighted and summed with the corresponding weight coefficients to generate a first comprehensive score for each of the tender documents. Based on the correlation index, a mapping matrix is ​​constructed between the scores of each indicator, including: when the correlation index between two target review indicators is greater than a preset threshold, calculating a mapping coefficient between the scores of the two target review indicators based on the correlation index, wherein the mapping coefficient represents the ideal proportional relationship between the scores of the two corresponding indicators; recording the mapping coefficient in the mapping matrix, wherein the rows and columns of the mapping matrix correspond to each review indicator, and the elements in the mapping matrix represent the mapping coefficient between the scores of the two review indicators corresponding to the corresponding row and column; when the correlation index between two target review indicators is less than or equal to the preset threshold, setting the corresponding mapping coefficient to a preset value, wherein the preset value indicates that there is no correlation between the two corresponding review indicators; the mapping matrix is ​​used to represent the proportional relationship between the scores of each indicator. Based on the mapping matrix, the standard score corresponding to each indicator score is determined. Based on the difference between each indicator score and the corresponding standard score, a first correction coefficient is generated for each bid document. Based on the first correction coefficient, the first comprehensive score is corrected to generate a second comprehensive score. Text similarity analysis is performed on the content of each bid document. Based on the text similarity analysis results, a second correction coefficient is generated for each bid document. The second comprehensive score is corrected based on the second correction coefficient to generate a target comprehensive score. The bid documents are then sorted based on the target comprehensive score to generate the review results.

2. The method for cross-validation of multiple documents based on format preservation according to claim 1, characterized in that, The calculation of the correlation index between each of the review indicators includes: obtaining preset business rules and review content of each of the review indicators; performing matching analysis on each of the review contents according to the business rules; when there is a logical relationship between any two of the review contents as defined by the business rules, calculating the business correlation index between the two review contents, wherein the business correlation index is positively correlated with the closeness of the logical relationship; and generating the correlation index between each of the review indicators based on the business correlation index, wherein the business correlation index is positively correlated with the correlation index.

3. The method for cross-validation of multiple documents based on format preservation according to claim 1, characterized in that, The step of generating a first correction coefficient for each bid document based on the difference between each indicator score and the corresponding standard score, and correcting the first comprehensive score based on the first correction coefficient to generate a second comprehensive score, includes: calculating the difference between each indicator score and the corresponding standard score; weighted summing of the absolute values ​​of each difference to generate a consistency deviation value for the bid document; calculating the first correction coefficient for each bid document based on the consistency deviation value, wherein the consistency deviation value is negatively correlated with the first correction coefficient; and arithmetically multiplying the first comprehensive score by the first correction coefficient to generate a second comprehensive score for the bid document.

4. The method for cross-validation of multiple documents based on format preservation according to claim 3, characterized in that, The step of calculating the first correction coefficient for each of the bid documents based on the consistency deviation value includes: obtaining a preset deviation benchmark value and a decay coefficient, substituting the consistency deviation value, the deviation benchmark value, and the decay coefficient into a preset formula to calculate the first correction coefficient for each of the bid documents; wherein, the preset formula is: C=max(L, 1-max(0, DB)×λ); where C is the first correction coefficient, D is the consistency deviation value, B is the deviation benchmark value, λ is the decay coefficient, and L is the preset lower limit value of the correction coefficient.

5. The method for cross-validation of multiple documents based on format preservation according to claim 1, characterized in that, The step of generating a second correction coefficient for each of the bid documents based on the text similarity analysis results, and then correcting the second comprehensive score based on the second correction coefficient to generate a target comprehensive score, includes: pairing each of the bid documents together, calculating the text similarity for each pair of bid documents to generate corresponding text similarity values, and preprocessing the images in each pair of bid documents to generate image feature sets; adjusting the text similarity values ​​based on the image feature sets to generate target text similarity values; generating a similarity aggregation index for each bid document based on each target text similarity value, wherein the similarity aggregation index represents the overall similarity between the bid document and other bid documents; calculating a second correction coefficient for each bid document based on the similarity aggregation index, wherein the similarity aggregation index is negatively correlated with the second correction coefficient; and arithmetically multiplying the second comprehensive score by the second correction coefficient to generate the target comprehensive score.

6. The method for cross-validation of multiple documents based on format preservation according to claim 5, characterized in that, The step of adjusting the text similarity value based on the image feature set to generate a target text similarity value includes: calculating the similarity value of the images in each group of bidding documents based on the image feature set, wherein the similarity value is used to characterize the degree of consistency between the images in two bidding documents; comparing the similarity value with a preset image similarity threshold, and marking the corresponding bidding document as having duplicate images when the similarity value is greater than the image similarity threshold; adjusting the text similarity value based on the marking result of duplicate images to generate a target text similarity value; wherein, when a bidding document is marked as having duplicate images, the target text similarity value is increased.

7. A multi-document cross-consistency verification system based on format preservation, characterized in that, The system includes: a first acquisition module, a second acquisition module, a first generation module, a construction module, a second generation module, and a third generation module; wherein, the first acquisition module is used to acquire a tender document, parse the tender document, generate multiple evaluation indicators, acquire the importance level of each evaluation indicator, and calculate the correlation index between each evaluation indicator; the second acquisition module is used to acquire multiple bid documents corresponding to the tender document, score each bid document through the multiple evaluation indicators, and generate indicator scores for each bid document corresponding to each evaluation indicator; the first generation module is used to generate weight coefficients for each evaluation indicator according to the importance level, and perform a weighted summation of each indicator score and the corresponding weight coefficient to generate a first comprehensive score for each bid document; the construction module is used to construct a mapping matrix between the indicator scores according to the correlation index, including: when the correlation index between two target evaluation indicators is greater than a preset threshold, calculating a mapping coefficient between the indicator scores corresponding to the two target evaluation indicators according to the correlation index, the mapping coefficient representing the ideal proportional relationship between the corresponding two indicator scores; and recording the mapping coefficient... The data is recorded in a mapping matrix, where the rows and columns of the mapping matrix correspond to each of the evaluation indicators, and the elements in the mapping matrix represent the mapping coefficient between the scores of the two evaluation indicators corresponding to the corresponding row and column. When the correlation index between two target evaluation indicators is less than or equal to the preset threshold, the corresponding mapping coefficient is set to a preset value, which indicates that there is no correlation between the two evaluation indicators. The mapping matrix is ​​used to characterize the proportional relationship between the scores of each indicator. The second generation module is used to determine the standard score corresponding to each indicator score according to the mapping matrix, generate a first correction coefficient for each bid document based on the difference between each indicator score and the corresponding standard score, correct the first comprehensive score according to the first correction coefficient, and generate a second comprehensive score. The third generation module is used to perform text similarity analysis on the content of each bid document, generate a second correction coefficient for each bid document based on the text similarity analysis results, correct the second comprehensive score according to the second correction coefficient, generate a target comprehensive score, sort the bid documents according to the target comprehensive score, and generate the evaluation result.

8. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1-6.

Images